A key opportunity of additive manufacturing (AM) is the ability to produce engineered structures whose design has been determined by simultaneously considering the shape and material complexity. Together, the shape and material couple to govern the attending properties, whether those properties are local (i.e., spatially varying) or global (i.e., load bearing capacity of a part). However, the tools lack the maturity and convergency to realize the dream of the simultaneous digital control of materials and shape for an optimized structure to achieve a desired performance level under dynamically varying external stimuli. The tools that are necessary include: AM; generative design and equivalents; gradient materials and heterogeneous materials; and agent-based artificial intelligent/machine learning protocols. In addition, while each of these tools requires additional investment and sustained development efforts, the metals AM community faces an additional barrier due to the fact that the current selection of materials available for use in AM is limited, and new alloys may be desired. To realize the discovery of new materials that may be produced using advanced manufacturing to manufacture advanced parts and components whose properties and performance transcend what is conceived of as “state-of-the-art,” it is necessary to transform the traditional laboratory ecosystem where advances are incremental and slow. One such concept is that of an autonomous laboratory for materials discovery. Within such an autonomous laboratory, an objective associated with a desirable shape or materials attribute (e.g., specific strength, electrical conductivity, magnetic susceptibility, …), including spatially varying complexities in shape and materials, would result in an AI agent and scientist collaborating to initiate (perhaps under some constraint) a series of fully integrated experiments involving materials composition, part shape, processing (synthesis), materials state, the corresponding properties, and the statistical nature of performance. These experiments would be tightly integrated into a dynamic and well-correlated multiscale digital thread that would provide the basis upon which physically relevant models would be determined and exercised, leading to rapid discovery of new materials as well as potentially the acceleration of new materials insertion, especially for additively manufactured components. While the details of such an autonomous laboratory where such articles are conceived, manufactured and tested might change, and include concepts such as physical or hybrid digital/physical test beds, the overall concepts remain the same. This article presents a concept of an autonomous laboratory, provides examples of building block advances in experimental and analytical techniques that could be leveraged to realize such laboratories, and describes some of the challenges, limitations, and opportunities of such a laboratory. The article also introduces the “Burke Test,” which is somewhat analogous to the Turing Test to assess AI, and provides metrics to determine whether a true autonomous laboratory has been achieved. Graphical abstract showing conceived autonomous laboratory. Left figure shows a nested image representing the six core sections of the scientific method and research development, each represented by a different color. S/W represents software, while Agt. represents agent or agents.
Collins et al. (Thu,) studied this question.